Reference to prior work

[ExpectationMax]

Hey Everybody,

first of all congrats to the paper, it's really interesting and am looking forward to seeing its impact on the space!

I wanted to point out that a few colleagues and me followed a similar approach to introduce translational equivariance in kernelizable attention (as is implemented in the Performer) for image classification tasks and posted it on ArXiv in the beginning of February https://arxiv.org/abs/2102.07680.

While the approach proposed in your work is more generic, we would highly appreciate if you could also refer to our prior work in the publication.

Best,
Max

[ZhuiyiTechnology]

Thank you for informing us of the related work. We will refer to your paper in the final version.